Reading time: 12 minutes
An attribution model is the rule that decides which channel gets credit for a sale. In GA4, the default attribution model is data-driven. But an AI conversation that never produces a click stays out of the count. So marketing attribution needs other evidence, like a form question and brand searches.
In short
- GA4 has three attribution models: data-driven, paid and organic last click, and Google paid channels last click.
- The first click, linear, position-based and time decay models have not been available since November 2023.
- No model credits a direct visit, unless the whole path is direct. So someone who found the brand in ChatGPT and came back through Google becomes an organic search sale.
- To measure what the click doesn’t show, add a “how did you hear about us?” field, UTMs and brand searches.
What is an attribution model in marketing?
An attribution model in marketing is the rule that splits the credit for a conversion among a customer’s touchpoints. Examples are an ad, a search, an email and a social post. So it answers which channels brought in the sale. But each model splits credit in its own way, and the weight of each channel changes with the choice.
According to Google Analytics Help, an attribution model can be a rule, a set of rules or a data-driven algorithm. In practice, marketing attribution uses five families of methods:
| Method | How it gives credit | Where it lives |
|---|---|---|
| Last click | 100% to the last channel | GA4 and Google Ads |
| Data-driven | By the estimated effect of each touchpoint | GA4 and Google Ads |
| First click, linear and others | By fixed rules | Left GA4 in 2023 |
| Marketing mix modeling | By the link between spend and sales | Outside GA4 |
| Self-reported | By what the customer says | Form and CRM |
In an AI journey, this choice matters more. After all, the ChatGPT conversation happens outside the site and often without a click. So an attribution model that only reads clicks has no way to credit it. The rest of this guide shows what GA4 sees, what it misses and how to complete your marketing attribution.

Which attribution models does GA4 offer today?
GA4 has three models in its attribution reports: data-driven attribution, paid and organic last click, and Google paid channels last click. The attribution model in GA4 applies to key events and revenue. It covers reports that use event-scoped dimensions, such as Source and Medium.
According to the GA4 attribution settings page, a change of attribution model applies to past and future data. However, session and user dimensions don’t change, such as Session source. In other words, a report built on Session source shows the same thing under any model.
Likewise, the page defines the lookback window, that is, how long a touchpoint can receive credit. The default is 90 days for most key events and 30 days for first_visit and first_open.
What left GA4, and when?
In April 2023, Google announced a change in Google Ads Help. Data-driven attribution became the default in Google Ads and GA4. In addition, the first click, linear, time decay and position-based models would go away. Starting in June 2023, Google Ads would stop offering them to anyone not already using them. Then, in September, it would start switching the ones still using them.
According to GA4 Help, these models have not been available since November 2023. Google backed the change with its own data: less than 3% of Google Ads web conversions used these models. So to see the first touch, use the first user dimensions, such as First user medium.
Which attribution model should you use in GA4?
Use data-driven attribution as your GA4 attribution model in reports, because it is the default and splits credit across channels. Then compare it with paid and organic last click. For marketing attribution, the gap between the two shows which channels open the path and which only close the sale.
Google’s help explains how the data-driven model works. It uses machine learning to evaluate both converting and non-converting paths. Then it compares what happened with what could have happened. It gives credit based on how likely each touchpoint is to drive the key event. Also, each model is specific to each advertiser and each key event, and conversions can be reattributed for up to 7 days.
The key event attribution models report, under Advertising > Attribution > Attribution models, shows each channel’s credit in two models side by side. The same page has an important note: GA4 uses last click for all Google Ads conversions based on key events.
- Data-driven: for day-to-day reports and for leadership.
- Paid and organic last click: to compare with Google Ads and see who closes the sale.
- Google paid channels last click: only for teams that evaluate Google Ads campaigns.
Why doesn’t last click see the influence of AI?
Last click gives 100% of the credit to the last channel clicked before the conversion, and it ignores direct traffic. So it rewards whoever closes the sale. The AI conversation, on the other hand, tends to sit early in the journey. And when it produces no click, it never enters the path at all.
GA4 Help gives an example: in the path display, social, paid search and direct, paid search gets 100%. Now swap the channels for an AI journey. First, a person asks ChatGPT for recommendations and reads the brand name. Then, they search for the brand on Google and click the organic result. Finally, they come back days later, type the address and buy.
In our reading of the help rules, last click gives the whole sale to organic search. With data-driven attribution, organic search also keeps the credit, because the model only splits it among recorded touchpoints. But the ChatGPT conversation left no record, since there was no click.
Even an AI visit without a referrer loses the credit, because it lands in Direct. According to the help, all attribution models exclude direct visits from credit, unless the path consists entirely of direct visits. Therefore, switching models doesn’t solve the AI influence problem. What helps is adding other sources to your marketing attribution.
How does the journey go from an AI chat to Google and direct?
In the journey we use as an example, it all starts with a question to the assistant, which names brands and products. Then the person checks on Google, often by searching the brand name, or types the address. So the influence of AI shows up later, as a brand search, an organic search or a direct visit.
In the US, AI is already part of this research. The 2026 Global Digital Shopping Index, by PYMNTS Intelligence and commissioned by Visa Acceptance Solutions, surveyed consumers and merchants in three countries, the US among them, in March 2026. Among shoppers whose latest purchase was online, 46% in the US used an AI platform or a merchant’s AI tool (47% across the three countries). Across all three countries, 30% of shoppers who looked up product information used ChatGPT before their latest purchase, up from 2% two years earlier. The report does not break that ChatGPT figure out for the US.
The US edition of the series also shows the marketing attribution problem in retail. According to the US Playbook (August 2026), only 24% of merchants can identify a sale that came from an AI platform. The page does not give the merchant sample for the US. The page presents the report as coming from PYMNTS Intelligence and Visa Acceptance Solutions. Visa is a payments company and therefore has an interest in the topic.
Where do AI visits without a referrer show up in GA4?
AI visits with no referrer and no UTM land in the Direct channel. According to the GA4 default channel definitions, Direct is a visit that arrives through a saved link or a typed URL. A click with a referrer, by contrast, goes to AI Assistant, Referral or Unassigned.
| How the visit arrives | Channel in GA4 |
|---|---|
| Click in an assistant, with a referrer | AI Assistant or Referral |
| Link with incomplete UTMs | Unassigned |
| App with no referrer or UTM | Direct |
| Address typed after the chat | Direct |
| Click in AI Overviews or AI Mode | Organic Search |
Unassigned is the value GA4 uses when no channel rule applies. According to GA4 Help on UTMs, if the URL has any UTM, the source comes only from the UTMs. A missing parameter shows up as (not set). OpenAI, for its part, used to document that ChatGPT added utm_source=chatgpt.com to links in answers with search.
The step-by-step setup for the AI Assistant channel and a regex channel group is in our guide to AI traffic in GA4. For attribution, the point is different: even a Direct visit that came from AI loses the credit to the previous channel.
How do you measure what the attribution model doesn’t see?
Add three sources to GA4, because no attribution model credits a conversation without a click. First, the “how did you hear about us?” question records what the customer remembers. Second, UTMs clean up Direct. Finally, period comparisons link brand searches to presence in AI answers. Together, they complete your marketing attribution.
How do you use the “how did you hear about us?” question?
Add an open field to your form and your CRM, because a closed list rarely includes “ChatGPT” or “an AI”. Then sort the answers every month and count the ones that name assistants. The step-by-step process for the field, the categories and the analysis is in our guide to self-reported attribution.
The field depends on the customer’s memory, but it records what no tracking sees. So report these sales separately, as a range, and don’t add them to the GA4 channel.
Where should you use UTMs to separate AI?
Use UTMs on every link you control: newsletters, social posts, partners, PDF materials and QR codes. That way, the remaining Direct traffic is cleaner and points better to AI and word of mouth. However, you don’t control the UTMs on links that AI creates. So filter Session source by chatgpt.com and the other assistants.
How do you compare periods and brand searches?
Search Console has a branded and non-branded queries filter. According to Search Console Help, it isn’t available for sites with few impressions, and it can misclassify some queries. The filter’s data starts in March 2025.
Then compare, month by month, clicks on brand searches, direct visits and the brand’s presence in AI answers. You can measure presence with the protocol to measure brand mentions and citations in AI. If all three rise together, that is a signal, not proof. If organic falls while brand searches stay flat, the cause may be something else; see how to diagnose an organic traffic drop.
Incremental attribution model or MMM: is it worth it for a mid-sized company?
In our view, for most mid-sized companies, not yet. An incremental attribution model uses tests to measure what would not happen without the channel. Marketing mix modeling (MMM), on the other hand, estimates each channel’s effect from sales and spend over time. Both are marketing attribution methods that need a lot of data, as Google’s example below shows.
Meridian is an open source MMM built by Google, and you can calibrate it with geo tests. On its page about the amount of data needed, Google gives an example. With 12 media channels, 6 controls and 8 knots, a national model has 26 parameters, in a simplified count. Two years of weekly data add up to 104 data points, or four per parameter. Google says this sample is too small.
So our advice is to start with what fits your operation: data-driven attribution, the form question and period comparisons. MMM comes in when you have several paid channels, years of weekly history and enough spend variation for the model to learn.
How do you present AI credit to leadership?
Show the GA4 number as a floor and always say which attribution model gave the credit. Then show, separately, the sales where the customer mentioned AI and the trend in brand searches. That way, leadership sees what marketing attribution measures and what is only a signal.
- Sales attributed in GA4: by channel, in the data-driven model, with last click next to it.
- AI channel: sessions and key events from assistants, as a floor.
- Self-reported attribution: sales and leads where the customer mentioned AI, as a range.
- Signs of influence: brand searches, direct visits and presence in AI answers, month by month.
To turn these numbers into a financial return, see how to calculate SEO and GEO ROI. Likewise, the digital marketing KPIs guide shows which of these numbers become targets. For content, the same logic applies to content marketing ROI.
Frequently asked questions
Yes. Changing the model in the GA4 attribution settings applies to past and future data, in reports with event-scoped dimensions. Session and user dimensions, such as Session source, don’t change with the model.
GA4 uses last click for all Google Ads conversions based on key events. Only key events where Google Ads was the last non-direct click become conversions in Google Ads. So, to compare the numbers, choose paid and organic last click in GA4.
Only when the path to the conversion has nothing but direct visits. Otherwise, no GA4 attribution model gives credit to the Direct channel. So a direct visit that came from an AI conversation loses the credit to the previous channel.
Main sources
- Google Analytics, Get started with attribution (2026).
- Google Analytics, attribution settings and lookback window (2026).
- Google Ads, announcement that rules-based models are going away (2023).
- Google Analytics, key event attribution models report (2026).
- Google Analytics, default channel group definitions (2026).
- PYMNTS Intelligence, 2026 Global Digital Shopping Index, commissioned by Visa Acceptance Solutions (2026).
- Search Console, branded and non-branded queries filter (2026).
- Google Meridian, amount of data needed for MMM (2026).



